Containing Agent Chaos
summary
Argues that isolating every agent step inside containers turns flaky LLM workflows into reproducible, debuggable pipelines, and demos a prototype that wires cloud-code and MCP to ephemeral per-action containers so each edit and run is sandboxed, snapshotted into git-backed snapshots and tunneled to a local machine. Uses Dagger primitives, supports secrets (OnePassword), runs multiple agent experiments (Claude, Goose) in parallel, and mostly works onstage despite rough edges — repo: github.com/dagger/container-use.
AI-written summary · about
chapters (5)
- 0:00Introduction and Docker Con nostalgia
- 3:34The rise of coding agents
- 8:15Why containers matter for agents
- 10:45Demo of container use
- 22:25Open source announcement
related talks
39mBuilding Agentic AI Apps with Docker MCP & Cagent
42mCI/CD Reimagined: Building Portable, Composable Pipelines with Dagger
42mThe Self-Healing Pipeline: How Dagger + LLMs Are Making Traditional CI/CD Obsolete
37mCreating chaos in containers
26mFuture-Proof Your Platform’s CI/CD Without Developers Noticing a Thing
18mFostering Chaos Engineering Culture
31mNo More 'git commit -m "fix pipeline #42"' Thanks to Dagger
30mDagger at Scale: How Not to Stab Yourself with Your Pipelines